Parallel Implementation of Density Functional Theory Methods in the Quantum Interaction Computational Kernel Program.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• High-Performance Computing

Background:

• Accurate calculation of exchange correlation (XC) energies and gradients is crucial for density functional theory (DFT) methods.
• Existing computational approaches often face performance bottlenecks, limiting the size and complexity of treatable molecular systems.

Purpose of the Study:

• To implement and detail a graphics processing unit (GPU) capable XC scheme within the open-source QUantum Interaction Computational Kernel (QUICK) program.
• To evaluate the performance and accuracy of the GPU-accelerated XC method compared to traditional CPU implementations.

Main Methods:

• Integration of an octree-based numerical grid partitioning scheme for efficient data management.
• Development of GPU-enabled grid pruning, basis function prescreening, and primitive function prescreening.
• Implementation of fully GPU-capable algorithms for XC energy and gradient computations.

Main Results:

• The GPU implementation achieved significant speedups in double precision XC energy (60-80-fold) and gradient (140-500-fold) computations for small to medium molecular systems on a NVIDIA V100 GPU.
• The performance gains from a single V100 GPU substantially surpassed those of a 40-core parallel CPU.
• The GPU implementation maintained excellent accuracy compared to the CPU version.

Conclusions:

• The developed GPU-accelerated XC scheme offers a highly efficient and accurate approach for electronic structure calculations.
• This advancement in computational chemistry can accelerate research in areas such as drug discovery and materials science by enabling the study of larger and more complex systems.